Prague AI Declaration

Overview

The Prague Declaration on Artificial Intelligence is a political and non-binding declaration adopted in Prague on 3 September 2026 by the governments of Croatia, the Czech Republic, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, and Slovenia. It was agreed at the CEE AI Summit and is expressly framed as a voluntary instrument that creates no legal obligations, no new institution, and no new budget line. Its purpose is to deepen regional cooperation on artificial intelligence while strengthening Central and Eastern Europe’s role in the wider European AI agenda. The declaration links competitiveness, technological sovereignty, security, resilience, and economic growth, while affirming that AI should be developed and deployed in line with European values, respect for human rights, personal data protection, safety, and clear human responsibility.

The document is organized around five sections. The first promotes know-how, talent, and mutual support; the second focuses on shared AI infrastructure; the third on coordination of European AI and digital policy; the fourth on a coherent and simplified regulatory framework for sovereign AI development; and the fifth on continued cooperation. Across these sections, the governments commit to practical collaboration on AI literacy, education and skills, public administration, healthcare, industry, research capacity, AI Factories and AI Gigafactories, regulatory sandboxes, coordinated positions in EU fora, and secure-by-design AI systems. The declaration is designed to operate within the participants’ competences and available resources, and it emphasizes that cooperation should be pragmatic and complementary to formal EU mechanisms rather than a substitute for them.

Definitions

The declaration does not contain a standalone definitions article comparable to a legislative act, but it uses several technical and policy terms in a consistent way. “AI Factories” and “AI Gigafactories” are treated as shared computing and infrastructure arrangements supporting the development, deployment, and scaling of AI models. “AI literacy” refers to the spread of practical understanding across society, including education and skills, teacher professional development, responsible use by learners, and evaluation of AI’s effects on teaching and learning. The declaration also refers to “regulatory sandboxes,” meaning controlled environments for testing innovative AI applications, and to “testing and evaluation facilities,” which are intended to support experimentation, validation, and deployment of AI solutions of common interest.

Other core concepts are “human responsibility,” “human oversight,” “fundamental rights,” “personal data,” “safety,” “security,” and “strategic dependencies.” The text uses these terms to mark the limits of automation and to insist that policy choices remain grounded in legal certainty and protection of rights. It also refers to “critical sectors,” explicitly including healthcare, and to “AI-enabled cyberattacks” and other malicious uses. The declaration’s vocabulary is therefore functional rather than definitional: it identifies priority domains and governance concerns, but it does not prescribe formal legal definitions, thresholds, or technical classifications. For practical use, the key interpretive point is that the instrument is cooperative and programmatic, not regulatory, and its terminology should be read in light of EU frameworks that already govern data protection, electronic health data, AI, and EuroHPC-related infrastructure.

Governance and Institutional Framework

Governance under the declaration is intentionally light and intergovernmental. The participating governments commit to ongoing coordination on AI and digital policy, including early consultation before meetings of the European Council, the Council of the EU, and the AI Board. They also express willingness to establish an informal network of ministerial and expert contact points. That network is meant to complement, not replace, formal EU procedures and national coordination mechanisms. It would support early notice of national initiatives and coordinated positions where appropriate and feasible, but it does not create a new supervisory authority or a regional bureaucracy.

The summit mechanism is also institutionalized only in a minimal way. The declaration states that the CEE AI Summit will convene annually in one of the Central and Eastern European countries, with the host setting the agenda and reporting on progress under the declaration, subject to available resources. This creates a recurring political forum for follow-up, but not a binding compliance body. The declaration repeatedly stresses that cooperation occurs within each participant’s competences, and it preserves national freedom to act in EU calls and competitive procedures. In institutional terms, the framework is therefore based on shared political commitments, informal coordination, and voluntary project cooperation rather than enforceable governance structures.

Key Focus Areas

The first major focus area is know-how, talent, and mutual support. The governments want public administrations to lead by example as responsible and secure adopters of AI, while positioning Central and Eastern Europe as a European leader in trustworthy, human-centred AI adoption. They also commit to advancing AI literacy across society, cooperating on education and skills, and sharing reusable solutions, procurement and evaluation tools, training programmes, AI literacy initiatives, and experts across four priority areas: education and skills, public administration, healthcare, and industry. Balanced territorial development is also part of this agenda, so AI benefits should reach regions, cities, and rural communities alike.

The second and third focus areas are shared AI infrastructure and policy coordination. The declaration supports mutual support for AI Factories and AI Gigafactory consortia, complementarity and interoperability between infrastructures, joint work on advanced AI models, and access to high-quality datasets in compliance with data protection and electronic health data rules. It also calls for deeper coordination on European AI and digital policy, including shared positions ahead of EU-level discussions. The fourth focus area is a coherent and simplified regulatory framework, with support for simplification of the EU digital framework, practical implementation of the AI Act, avoidance of unjustified additional national requirements, and retention of a high level of protection for fundamental rights, personal data, safety, and security. The fifth focus area is continued cooperation on AI security, cybersecurity, critical infrastructure resilience, and the detection and mitigation of malicious uses of AI, including AI-enabled information manipulation.

Implementation Framework

Implementation is framed as voluntary cooperation delivered through national competences, available resources, and practical project-based work. The declaration identifies specific instruments for implementation, including reusable solutions, procurement tools, evaluation tools, training programmes, shared expert support, cross-border regulatory sandboxes, joint pilots, coordinated public procurement, shared testing and evaluation facilities, and a possible common framework linking the region’s AI factories and gigafactories. These mechanisms are presented as explorations or commitments to cooperate, not as mandatory legal requirements. The declaration also points to cooperation on access to compute and finance for start-ups and spin-offs, as well as efforts to attract and retain researchers, founders, and other talent in the region.

In infrastructure terms, the declaration seeks interoperability and mutual access to compute, while remaining in full respect of EuroHPC governance and national participation in competitive calls. In policy terms, it encourages early consultation and coordinated positions in European decision-making. In regulatory terms, it asks participants to avoid unjustified additional national requirements that could fragment the market, while taking into account national supervisory, testing, and conformity assessment capacities. The overall implementation model is therefore modular and adaptive: it combines strategic alignment at political level with practical cooperation at technical and administrative levels, and it leaves each government free to decide how far to act domestically. This is consistent with the declaration’s explicit statement that it does not create legal obligations.

Monitoring and Evaluation

Monitoring is built around the annual CEE AI Summit. The declaration provides that the summit will convene each year in one of the participating countries, and the host will set the agenda and report on progress under the declaration, subject to available resources. That reporting function is the main evaluation mechanism identified in the text. It is meant to track how cooperation advances on infrastructure, policy coordination, talent, public administration, healthcare, and industrial adoption, but the declaration does not prescribe indicators, benchmarks, or a formal review template. As a result, monitoring is political and operational rather than legal or statistical.

The declaration also implies a softer form of evaluation through the sharing of practice, pilots, and tools. It endorses evidence-based evaluation of AI’s impact on teaching and learning, shared testing and evaluation facilities, and practical assistance across administrations and sectors. These elements suggest that performance will be judged by usefulness, interoperability, and uptake, not by formal compliance scoring. The text does not establish a secretariat, audit mechanism, or enforcement body. Instead, progress is expected to be assessed through follow-up at successive summits and through cooperation within existing national and EU structures. That makes the monitoring model deliberately lightweight, suitable for a non-binding declaration whose objective is coordination rather than sanction.

Penalties, Liability, and Appeals

The declaration contains no penalties, fine regime, or criminal sanctions, and it creates no new liability rules. It expressly states that it is a political and non-binding document that creates no legal obligations, no new institution, and no new budget line. Cooperation is voluntary and occurs within the participants’ competences and available resources. It is also without prejudice to the rights and obligations of the participants. Because the instrument does not impose binding duties, there is no built-in appeal process, no dispute settlement clause, and no enforcement architecture comparable to a treaty or regulation.

Any issues arising from the declaration would therefore remain within the relevant national or EU legal frameworks, such as those governing data protection, AI regulation, procurement, health data, cybersecurity, or administrative law. The declaration does not alter liability allocation for AI-related harms, nor does it provide compensation mechanisms for individuals or businesses. It does, however, reaffirm the need for clear human responsibility and appropriate human oversight where AI may have significant consequences for individuals, society, or security. That statement is policy-oriented and does not itself create a cause of action. In practical terms, the document is best understood as a coordination pledge backed by political accountability and public reporting, not legal remedies or sanctions.

Relationship to Other Instruments

The declaration situates itself alongside, and subordinate to, existing European frameworks. It explicitly refers to the AI Act and calls for its implementation to be practical, proportionate, and consistent across the Single Market. It also welcomes the Digital Omnibus, including the Digital Omnibus on AI, as part of the effort to simplify the EU digital framework, and it supports the European Technological Sovereignty Package as a framework for cloud and AI infrastructure, open-source technologies, and sustainable digital infrastructure. In healthcare, it notes the importance of reliable and interoperable data infrastructure and refers to Regulation (EU) 2025/327 on the European Health Data Space.

The text also emphasizes that cooperation must be fully aligned with EuroHPC governance and compatible with ongoing competitive procedures and each country’s freedom to compete in European calls. It further states that the informal ministerial and expert network would complement, not replace, formal EU procedures and national coordination mechanisms. The declaration therefore operates as a political layer above existing law, intended to improve alignment and reduce fragmentation without changing binding obligations. It is not a substitute for national AI strategies, EU legislation, or sectoral rules. Instead, it is a cross-border cooperation instrument that seeks to make the region an active contributor to the European AI and technological agenda while preserving the legal primacy of applicable EU and national instruments.

International Alignment

International alignment is central to the declaration’s rationale. The participating governments state that Central and Eastern Europe should strengthen its contribution to a competitive, innovative, technologically advanced, and sovereign Europe grounded in European values and principles. They emphasize that Europe’s technological strength must combine stronger European capabilities with access to the best globally available technologies, expertise, infrastructure, and investment. The declaration also insists that policy choices should be guided by competitiveness, security, resilience, and economic growth, linking regional cooperation to the broader European effort to scale AI technologies.

At the same time, the declaration seeks to reduce strategic dependencies and protect Europe’s ability to create and steer its own AI capability. It supports cooperation on compute, models, data, and infrastructure, but in a way that respects EU law and market mechanisms. The text repeatedly references European-level coordination, especially around the Council of the EU, the European Council, the AI Board, EuroHPC, and the AI Act. It also highlights the need to support AI models that work reliably in less widely spoken regional languages, which is a practical contribution to European technological inclusion. Overall, the declaration is aligned with European integration and digital sovereignty goals, while remaining explicitly non-binding and nationally voluntary.

Implementation Timeline

Compliance Checklist

Sources and References